Abstract: Aiming at the problems of low recognition accuracy for compound jamming modes and insufficient generalization ability of traditional methods in modern electronic warfare environments, this paper proposes a radar jamming style recognition method based on time-frequency joint analysis and a lightweight convolutional neural network (CNN). The key innovations include: 1) constructing a large-scale dataset covering 9 basic jamming modes and 120 compound jamming combinations, and generating time-frequency joint feature images by vertically concatenating time-domain waveforms and frequency-domain spectra to fully exploit complementary information; 2) designing a lightweight CNN model with only 0.23M parameters, employing hierarchical convolution and extreme dimensionality reduction in the fully connected layer to significantly reduce computational complexity while maintaining high accuracy, enabling efficient deployment on CPUs. Experimental results show that the proposed method achieves a recognition accuracy of over 98% for single jamming modes and maintains 92.1% accuracy for triple-compound jamming scenarios, with an overall average recognition accuracy of 96.3%. Moreover, it enables efficient training and recognition without GPU support. This study provides reliable data support and a model foundation for intelligent radar counter-countermeasure systems in complex electromagnetic environments, demonstrating good prospects for engineering applications.